The rise of the Deep Neural Network as an increasingly universal paradigm of computation has been the defining feature across much of computing in the last few years. At the same time, the traditional von Neumann processor paradigm is being challenged by the rise of hardware spatial computational accelerators, such as FPGAs, which face serious usability and programmability challenges. The deep learning application domain is narrow enough to allow us to reconsider the entire stack of spatial compute, from arithmetic circuits to cloud-based systems, in an integrated and domain-specific manner. We have strong expertise across the breadth of this space. If successful, our joint research centre has the potential to put the UK's expertise at the centre of the technology revolutionising the way high performance and low energy computation is specified and delivered, opening opportunities from ultra-low energy machine learning in internet of things devices through to powering scientific discovery in high-end server farms. This centre would mark a break-through in international coordination of research in the field. We will: (a) work together to evolve a joint research strategy, (b) deliver elements of that joint research strategy through the staff employed on this grant as well as through academic staff and PhD students in all institutions, (c) facilitate a managed researcher exchange programme through funding exchange / secondment expenses for investigators and visiting researchers, as well as supporting secondments to/from industry (d) hold annual showcase events of our work, and (e) deliver high-quality R&D and STEM advocacy outreach. By the end of the three-year period, we expect to have a self-sustaining momentum of internationally-coordinated research, incorporating the initial investigators, but extending beyond to new groups and the network of SMEs developing in this area.

Cancer is the second most common cause of death globally, accounting for 8.8 million deaths in 2015. It is estimated that radiotherapy is used in the treatment of approximately half of all cancer patients. In the UK, one new NHS proton-beam therapy facility has recently come online in Manchester and a second will soon be brought into operation in London. In addition, several new private proton-beam therapy facilities are being developed. The use of these new centres, and the research that will be carried out to enhance the efficacy of the treatments they deliver, will substantially increase demand. Worldwide interest in particle-beam therapy (PBT) is growing and a significant growth in demand in this technology is anticipated. By 2035, 26.9 million life-years in low- and middle-income countries could be saved if radiotherapy capacity could be scaled up. The investment required for this expansion will generate substantial economic gains. Radiotherapy delivered using X-ray beams or radioactive sources is an established form of treatment widely exploited to treat cancer. Modern X-ray therapy machines allow the dose to be concentrated over the tumour volume. X-ray dose falls exponentially with depth so that the location of primary tumours in relation to heart, lungs, oesophagus and spine limits dose intensity in a significant proportion of cases. The proximity of healthy organs to important primary cancer sites implies a fundamental limit on the photon-dose intensities that may be delivered. Proton and ion beams lose the bulk of their energy as they come to rest. The energy-loss distribution therefore has a pronounced 'Bragg peak' at the maximum range. Proton and ion beams overcome the fundamental limitation of X-ray therapy because, in comparison to photons, there is little (ions) or no (protons) dose deposited beyond the distal tumour edge. This saves a factor of 2-3 in integrated patient dose. In addition, as the Bragg peak occurs at the maximum range of the beam, treatment can be conformed to the tumour volume. Protons with energies between 10MeV and 250MeV can be delivered using cyclotrons which can be obtained `off the shelf' from a number of suppliers. Today, cyclotrons are most commonly used for proton-beam therapy. Such machines are not able to deliver multiple ion species over the range of energies required for treatment. Synchrotrons are the second most common type of accelerator used for proton- and ion-beam therapy and are more flexible than cyclotrons in the range of beam energy that can be delivered. However, the footprint, complexity and maintenance requirements are all larger for synchrotrons than for cyclotrons, which increases the necessary investment and the running costs. We propose to lay the technological foundations for the development of an automated, adaptive system required to deliver personalised proton- and ion-beam therapy by implementing a novel laser-driven hybrid accelerator system dedicated to the study of radiobiology. Over the two years of this programme we will: * Deliver an outline CDR for the 'Laser-hybrid Accelerator for Radiobiological Applications', LhARA; * Establish a test-bed for advanced technologies for radiobiology and clinical radiotherapy at the Clatterbridge Cancer Centre; and * Create a broad, multi-disciplinary UK coalition, working within the international Biophysics Collaboration to place the UK in pole position to contribute to, and to benefit from, this exciting new biomedical science-and-innovation initiative.

The DART project aims to pioneer a ground-breaking capability to enhance the performance and energy efficiency of reconfigurable hardware accelerators for next-generation computing systems. This capability will be achieved by a novel foundation for a transformation engine based on heterogeneous graphs for design optimisation and diagnosis. While hardware designers are familiar with transformations by Boolean algebra, the proposed research promotes a design-by-transformation style by providing, for the first time, tools which facilitate experimentation with design transformations and their regulation by meta-programming. These tools will cover design space exploration based on machine learning, and end-to-end tool chains mapping designs captured in multiple source languages to heterogeneous reconfigurable devices targeting cloud computing, Internet-of-Things and supercomputing. The proposed approach will be evaluated through a variety of benchmarks involving hardware acceleration, and through codifying strategies for automating the search of neural architectures for hardware implementation with both high accuracy and high efficiency.